private void FastWrite(WriteOp op, List <SessionState> sessions = null, ServerMonitor monitor = null)
{
// there should only be one primary server since we are in fast mode
string server = configuration.PrimaryServers.First();
ICloudBlob blob = ClientRegistry.GetCloudBlob(server, configuration.Name, blobName, false);
watch.Start();
op(blob);
watch.Stop();
// update server and session state
ServerState ss = (monitor == null) ? slaEngine.Monitor.GetServerState(server) : monitor.GetServerState(server);
ss.AddRtt(watch.ElapsedMilliseconds);
if (sessions == null)
{
slaEngine.Session.RecordObjectWritten(blobName, Timestamp(blob), ss);
}
else
{
foreach (SessionState session in sessions)
{
session.RecordObjectWritten(blobName, Timestamp(blob), ss);
}
}
}
/// <summary>
/// Sets the WiP flags on all of the primary blobs.
/// </summary>
/// <returns>a dictionary containing the returned eTags for each primary;
/// returns null if the flag-setting protocol was not successful</returns>
private Dictionary <string, string> SetWiPFlags()
{
ICloudBlob blob;
Dictionary <string, string> eTags = new Dictionary <string, string>();
bool didAtLeastOne = false;
foreach (string server in configuration.PrimaryServers)
{
blob = ClientRegistry.GetCloudBlob(server, configuration.Name, blobName, false);
blob.Metadata[ConstPool.WRITE_IN_PROGRESS] = ConstPool.WRITE_IN_PROGRESS;
try
{
blob.SetMetadata();
didAtLeastOne = true;
}
catch (StorageException)
{
// If setting the flag fails at some primary, then abort the protocol
// Note that some WiP flags may have already been set, so we first try to clear them
if (didAtLeastOne)
{
ClearWiPFlags();
}
return(null);
}
eTags[server] = blob.Properties.ETag;
}
return(eTags);
}
private void DoDelete(Microsoft.WindowsAzure.Storage.Blob.DeleteSnapshotsOption deleteSnapshotsOption = DeleteSnapshotsOption.None, Microsoft.WindowsAzure.Storage.AccessCondition accessCondition = null, Microsoft.WindowsAzure.Storage.Blob.BlobRequestOptions options = null, Microsoft.WindowsAzure.Storage.OperationContext operationContext = null)
{
bool done = false;
while (!done)
{
using (PrimaryCloudBlobLease lease = new PrimaryCloudBlobLease(this.Name, configuration, true))
{
if (lease.HasLease)
{
Dictionary <ICloudBlob, IAsyncResult> results = new Dictionary <ICloudBlob, IAsyncResult>();
foreach (string serverName in configuration.PrimaryServers)
{
watch.Start();
ICloudBlob blob = ClientRegistry.GetCloudBlob(serverName, configuration.Name, Name);
results[blob] = blob.BeginDelete(deleteSnapshotsOption, lease.getAccessConditionWithLeaseId(accessCondition), options, operationContext, null, null);
ServerState ss = slaEngine.Monitor.GetServerState(serverName);
ss.AddRtt(watch.ElapsedMilliseconds);
slaEngine.Session.RecordObjectWritten(Name, Timestamp(blob), ss);
}
foreach (ICloudBlob blob in results.Keys)
{
blob.EndDelete(results[blob]);
}
done = true;
}
}
}
}
/// <summary>
/// Returns the blob for the main primary site of the current configuration.
/// </summary>
/// <returns>the primary blob (or null if unable to renew the lease on the configuration)</returns>
public ICloudBlob MainPrimary()
{
ICloudBlob primary = null;
if (configuration.IsInFastMode(renew: true))
{
string server = configuration.PrimaryServers.First();
primary = ClientRegistry.GetCloudBlob(server, configuration.Name, blobName);
}
return(primary);
}
/// <summary>
/// Perform standard read and update server and session state.
/// </summary>
/// <param name="op">The read operation</param>
/// <param name="blob">The blob being read</param>
/// <param name="ss">The chosen server</param>
private void FastRead(ReadOp op)
{
ServerState ss = slaEngine.FindServerToRead(blobName);
blobForRead = ClientRegistry.GetCloudBlob(ss.Name, containerName, blobName);
watch.Start();
op(blobForRead);
watch.Stop();
slaEngine.RecordObjectRead(blobForRead.Name, Timestamp(blobForRead), ss, watch.ElapsedMilliseconds);
}
public void FetchAttributes(Microsoft.WindowsAzure.Storage.AccessCondition accessCondition = null, Microsoft.WindowsAzure.Storage.Blob.BlobRequestOptions options = null, Microsoft.WindowsAzure.Storage.OperationContext operationContext = null)
{
// TODO: use protocol
ServerState ss = slaEngine.FindServerToRead(Name);
ICloudBlob blob = ClientRegistry.GetCloudBlob(ss.Name, configuration.Name, Name);
watch.Start();
blob.FetchAttributes(accessCondition, options, operationContext);
ss.AddRtt(watch.ElapsedMilliseconds);
// slaEngine.SessionState.RecordObjectRead(blob.Name, Timestamp(blob), ss);
slaEngine.Session.RecordObjectRead(blob.Name, Timestamp(blob), ss, "");
}
/// <summary>
/// Perform a local write operation, e.g. set a property, on all blobs holding a replica.
/// This does not bother to check the lease on the configuration.
/// It should not be used for arbitrary writes that need to be performed atomically.
/// </summary>
/// <param name="op">the write operation being done</param>
public void SetProperty(WriteOp op)
{
foreach (string server in configuration.PrimaryServers)
{
ICloudBlob blob = ClientRegistry.GetCloudBlob(server, configuration.Name, blobName);
op(blob);
}
foreach (string server in configuration.SecondaryServers)
{
ICloudBlob blob = ClientRegistry.GetCloudBlob(server, configuration.Name, blobName);
op(blob);
}
}
/// <summary>
/// Lease configuration container responsible for this blob along with blobs in the containers
/// </summary>
/// <param name="blobName">Particular blobs we would like to lease</param>
/// <param name="containerElementSet">list of containers</param>
/// <param name="putOptimization">perform put optimization. I.e., for put operations, it is not required to take lease if there is only one primary.</param>
public PrimaryCloudBlobLease(string blobName, ReplicaConfiguration configuration, bool putOptimization = false)
{
if ((putOptimization) && configuration.PrimaryServers.Count == 1)
{
//There is only one primary. No need to
HasLease = true;
return;
}
leasedBlobs = new Dictionary <ICloudBlob, string>();
//Reconfiguration is not going to happen in near future.
//We can safely take leases of primary blobs.
this.ProposedLeaseId = Guid.NewGuid().ToString();
foreach (string serverName in configuration.PrimaryServers)
{
try
{
ICloudBlob blob = ClientRegistry.GetCloudBlob(serverName, configuration.Name, blobName);
if (!blob.Exists())
{
//we cannot take lease on a non-existing blob.
//Hence, we create it first.
byte[] dummy = new byte[1];
dummy[0] = (byte)0;
var ms = new MemoryStream(dummy);
blob.UploadFromStream(ms, null, null, null);
}
string leaseID = blob.AcquireLease(null, ProposedLeaseId);
leasedBlobs.Add(blob, leaseID);
}
catch (StorageException ex)
{
releaseContainers();
leasedBlobs.Clear();
HasLease = false;
if (ex.GetBaseException().Message.Contains("409")) //Conflict
{
return;
}
else
{
throw ex;
}
}
}
HasLease = true;
}
/// <summary>
/// Clears the WiP flags on all of the primary blobs.
/// </summary>
/// <param name="eTags">if not null, then the flags should be cleared conditionally</param>
private void ClearWiPFlags(Dictionary <string, string> eTags = null)
{
ICloudBlob blob;
AccessCondition access = AccessCondition.GenerateEmptyCondition();
// Clear WiP flags at non-main primaries first
foreach (string server in configuration.PrimaryServers.Skip(1))
{
blob = ClientRegistry.GetCloudBlob(server, configuration.Name, blobName, false);
blob.Metadata.Remove(ConstPool.WRITE_IN_PROGRESS);
if (eTags != null)
{
access = AccessCondition.GenerateIfMatchCondition(eTags[server]);
}
try
{
blob.SetMetadata(access);
}
catch (StorageException)
{
// Ignore failures since the only consequence is that the Wip flag remains set
// It could be that another write is still in progress
// The flag will be cleared eventually by another writer or by the recovery process
}
}
// Clear WiP on main primary last
blob = ClientRegistry.GetCloudBlob(configuration.PrimaryServers.First(), configuration.Name, blobName, false);
blob.Metadata.Remove(ConstPool.WRITE_IN_PROGRESS);
if (eTags != null)
{
access = AccessCondition.GenerateIfMatchCondition(eTags[configuration.PrimaryServers.First()]);
}
try
{
blob.SetMetadata(access);
}
catch (StorageException)
{
// Ignore
}
}
private void MultiWriteUsingBlobLease(WriteOp op)
{
// TODO: recover from failed clients that may leave a write partially completed
// TODO: remove use of leases and replace with ETags
// throw new Exception("Write to multiple primaries not yet implemented.");
try
{
bool done = false;
while (!done)
{
// grab lease on blob to guard against concurrent writers
using (PrimaryCloudBlobLease lease = new PrimaryCloudBlobLease(blobName, configuration, true))
{
if (lease.HasLease)
{
// TODO: fix code for writing to multiple primaries (and remove it from here)
foreach (string server in configuration.PrimaryServers)
{
watch.Start();
ICloudBlob blob = ClientRegistry.GetCloudBlob(server, configuration.Name, blobName, false);
op(blob);
watch.Stop();
ServerState ss = slaEngine.Monitor.GetServerState(server);
ss.AddRtt(watch.ElapsedMilliseconds);
slaEngine.Session.RecordObjectWritten(blobName, Timestamp(blob), ss);
}
done = true;
}
}
}
}
catch (StorageException se)
{
throw se;
}
catch (Exception ex)
{
throw ex;
}
}
private void DoUploadFromStream(System.IO.Stream source, Microsoft.WindowsAzure.Storage.AccessCondition accessCondition = null, Microsoft.WindowsAzure.Storage.Blob.BlobRequestOptions options = null, Microsoft.WindowsAzure.Storage.OperationContext operationContext = null)
{
try
{
bool done = false;
while (!done)
{
using (PrimaryCloudBlobLease lease = new PrimaryCloudBlobLease(Name, configuration, true))
{
if (lease.HasLease)
{
foreach (string server in configuration.PrimaryServers)
{
watch.Start();
ICloudBlob blob = ClientRegistry.GetCloudBlob(server, configuration.Name, Name, false);
source.Position = 0;
blob.UploadFromStream(source, lease.getAccessConditionWithLeaseId(accessCondition), options, operationContext);
watch.Stop();
ServerState ss = slaEngine.Monitor.GetServerState(server);
ss.AddRtt(watch.ElapsedMilliseconds);
slaEngine.Session.RecordObjectWritten(Name, Timestamp(blob), ss);
}
done = true;
}
}
}
}
catch (StorageException se)
{
throw se;
}
catch (Exception ex)
{
throw ex;
}
}
private void MultiWrite(WriteOp op, AccessCondition access, List <SessionState> sessions = null, ServerMonitor monitor = null)
{
// This protocol uses three phases and ETags
// It assumes that the client is in fast mode and remains so throughout the protocol
// i.e. it assumes that the set of primaries does not change.
// TODO: recover from failed clients that may leave a write partially completed
ICloudBlob blob;
Dictionary <string, string> eTags;
// Phase 1. Mark intention to write with write-in-progress flags
eTags = SetWiPFlags();
if (eTags == null)
{
// flags were not successfully set, so abort the protocol
return;
}
// Phase 2. Perform write at all primaries
bool didAtLeastOne = false;
foreach (string server in configuration.PrimaryServers)
{
blob = ClientRegistry.GetCloudBlob(server, configuration.Name, blobName, false);
access.IfMatchETag = eTags[server];
watch.Start();
try
{
op(blob);
}
catch (StorageException)
{
// If writing fails at some primary, then abort the protocol
// It could be that a concurrent writer is in progress
// Note that some writes may have already been performed
// If so, then we leave the WiP flags set so the recovery process will kick in
// or so a concurrent writer can complete its protocol and overwrite our writes
// If not, then we can clear the WiP flags
if (!didAtLeastOne)
{
ClearWiPFlags(eTags);
}
return;
}
watch.Stop();
eTags[server] = blob.Properties.ETag;
didAtLeastOne = true;
// update session and server state
ServerState ss = (monitor == null) ? slaEngine.Monitor.GetServerState(server) : monitor.GetServerState(server);
ss.AddRtt(watch.ElapsedMilliseconds);
if (sessions == null)
{
slaEngine.Session.RecordObjectWritten(blobName, Timestamp(blob), ss);
}
else
{
foreach (SessionState session in sessions)
{
session.RecordObjectWritten(blobName, Timestamp(blob), ss);
}
}
}
// Phase 3. Clear write-in-progress flags to indicate that write has completed
ClearWiPFlags(eTags);
}
/// <summary>
/// Perform read optimistically and then verify configuration.
/// </summary>
/// <param name="op">The read operation</param>
/// <param name="blob">The blob being read</param>
/// <param name="ss">The chosen server</param>
/// <returns>whether the read succeeded; if not, then it should be tried again.</returns>
private void SlowRead(ReadOp op)
{
// TODO: Deal with the case that we try to read from a replica that is no longer a replica and the read fails
// In this case, the read should be retried after refreshing the configuration.
ServerState ss = null;
try
{
bool isDone = false;
do // until we enter fast mode or succeed at reading in slow mode with the correct configuration
{
ss = slaEngine.FindServerToRead(blobName);
blobForRead = ClientRegistry.GetCloudBlob(ss.Name, containerName, blobName);
// TODO: this should really check if the reader wants strong consistency
// It could be that eventual consistency is fine but the SLA Engine just happened to choose a primary server
// In this case, we can do a fast mode read since we don't care if the chosen server is no longer a primary
if (configuration.IsInFastMode() || !configuration.PrimaryServers.Contains(ss.Name))
{
// it is fine to read from the selected secondary replica
// or from a primary replica if we have now entered fast mode
FastRead(op);
isDone = true;
}
else
{
// read from the selected replica that we believe to be a primary replica
watch.Start();
op(blobForRead);
watch.Stop();
// then see if the configuration has changed and the selected replica is no longer primary
configuration.SyncWithCloud(ClientRegistry.GetConfigurationAccount());
// TODO: check the epoch number on the configuration to make sure that we have not missed a configuration.
// i.e. we need to make sure that the server from which we read did not become a secondary during our read,
// and then back to primary when we checked.
// TODO: maybe we should check that the read delivered a non-zero utility.
// It is possible that the configuration was so stale that a much better server could have been chosen.
if (configuration.PrimaryServers.Contains(ss.Name))
{
//We have contacted the primary replica, hence we are good to go.
slaEngine.RecordObjectRead(blobForRead.Name, Timestamp(blobForRead), ss, watch.ElapsedMilliseconds);
isDone = true;
}
isDone = false; // not done
}
} while (!isDone);
}
// TODO: decide if we need to catch any exceptions here or just let then propagate through
catch (StorageException se)
{
if (StorageExceptionCode.NotFound(se))
{
//blob is not found.
//this is fine because the replica might have removed.
//it simply returns, so client need to re-execute if this happens.
//We can also re-execute the read here, but for debugging purposes, it's simpler for the client to re-execute.
//Of course in real system, we need to re-execute at this level.
return;
}
// storage exceptions are passed through to the caller
throw;
}
catch (Exception ex)
{
if (ex.Message.Contains("Object reference not set to an instance of an object"))
{
return;
}
throw ex;
}
}
